Hugging Face's logo

Datasets:
applied-ai-018
/
peacock-data-public-datasets-idc-bigscience

Dataset card Files Community
peacock-data-public-datasets-idc-bigscience / data /mc4 /README.md
applied-ai-018's picture
applied-ai-018
Add files using upload-large-folder tool
9605203 verified
preview code
|
raw
history blame contribute delete
1.04 kB

mc4

Megatron pre-processed files

These are the megatron-ready mc4 files:

  • 1.3TB: $six_ALL_CCFRWORK/datasets-custom/mc4/mc4_preprocessing

Should something get corrupted there is a backup:

  • 1.3TB: $six_ALL_CCFRSTORE/datasets-custom/mc4/mc4_preprocessing

If files need to re-pre-processed, the original jsonl files are at:

  • 186GB: $six_ALL_CCFRSTORE/datasets-custom/mc4/mc4_sampled_raw

How pre-processing was done

The pre-processing was done outside of JZ, and was downloaded from:

To download one needs to activate the already installed on JZ google-cloud-sdk and then use gsutil as instructed at the Download tab in the links above.